WO2022131138A1 - 推定装置、推定方法、推定プログラム、及び学習モデル生成装置 - Google Patents
推定装置、推定方法、推定プログラム、及び学習モデル生成装置 Download PDFInfo
- Publication number
- WO2022131138A1 WO2022131138A1 PCT/JP2021/045417 JP2021045417W WO2022131138A1 WO 2022131138 A1 WO2022131138 A1 WO 2022131138A1 JP 2021045417 W JP2021045417 W JP 2021045417W WO 2022131138 A1 WO2022131138 A1 WO 2022131138A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- flexible material
- estimation
- electrical characteristics
- shape
- learning model
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims description 47
- 239000000463 material Substances 0.000 claims abstract description 117
- 238000001514 detection method Methods 0.000 claims abstract description 72
- 230000008859 change Effects 0.000 claims abstract description 14
- 230000008569 process Effects 0.000 claims description 30
- 230000004044 response Effects 0.000 claims description 7
- JOYRKODLDBILNP-UHFFFAOYSA-N Ethyl urethane Chemical compound CCOC(N)=O JOYRKODLDBILNP-UHFFFAOYSA-N 0.000 claims description 6
- 238000012545 processing Methods 0.000 description 23
- 238000012795 verification Methods 0.000 description 15
- 230000006870 function Effects 0.000 description 12
- 230000000638 stimulation Effects 0.000 description 10
- 238000013528 artificial neural network Methods 0.000 description 8
- 238000012360 testing method Methods 0.000 description 6
- 238000012549 training Methods 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 5
- 238000004891 communication Methods 0.000 description 5
- 238000000513 principal component analysis Methods 0.000 description 4
- 230000006399 behavior Effects 0.000 description 3
- 238000013480 data collection Methods 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 239000004020 conductor Substances 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000006073 displacement reaction Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000005452 bending Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000003703 image analysis method Methods 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000002861 polymer material Substances 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B7/00—Measuring arrangements characterised by the use of electric or magnetic techniques
- G01B7/16—Measuring arrangements characterised by the use of electric or magnetic techniques for measuring the deformation in a solid, e.g. by resistance strain gauge
- G01B7/18—Measuring arrangements characterised by the use of electric or magnetic techniques for measuring the deformation in a solid, e.g. by resistance strain gauge using change in resistance
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B7/00—Measuring arrangements characterised by the use of electric or magnetic techniques
- G01B7/28—Measuring arrangements characterised by the use of electric or magnetic techniques for measuring contours or curvatures
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01L—MEASURING FORCE, STRESS, TORQUE, WORK, MECHANICAL POWER, MECHANICAL EFFICIENCY, OR FLUID PRESSURE
- G01L1/00—Measuring force or stress, in general
- G01L1/20—Measuring force or stress, in general by measuring variations in ohmic resistance of solid materials or of electrically-conductive fluids; by making use of electrokinetic cells, i.e. liquid-containing cells wherein an electrical potential is produced or varied upon the application of stress
- G01L1/22—Measuring force or stress, in general by measuring variations in ohmic resistance of solid materials or of electrically-conductive fluids; by making use of electrokinetic cells, i.e. liquid-containing cells wherein an electrical potential is produced or varied upon the application of stress using resistance strain gauges
- G01L1/225—Measuring circuits therefor
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01L—MEASURING FORCE, STRESS, TORQUE, WORK, MECHANICAL POWER, MECHANICAL EFFICIENCY, OR FLUID PRESSURE
- G01L1/00—Measuring force or stress, in general
- G01L1/20—Measuring force or stress, in general by measuring variations in ohmic resistance of solid materials or of electrically-conductive fluids; by making use of electrokinetic cells, i.e. liquid-containing cells wherein an electrical potential is produced or varied upon the application of stress
- G01L1/22—Measuring force or stress, in general by measuring variations in ohmic resistance of solid materials or of electrically-conductive fluids; by making use of electrokinetic cells, i.e. liquid-containing cells wherein an electrical potential is produced or varied upon the application of stress using resistance strain gauges
- G01L1/2268—Arrangements for correcting or for compensating unwanted effects
Definitions
- the present disclosure relates to an estimation device, an estimation method, an estimation program, and a learning model generation device.
- the system including the camera and image analysis becomes a large scale. This is not preferable because it causes an increase in the size of the device.
- the techniques of the present disclosure include a detector that detects electrical properties between a plurality of detection points in a conductive flexible material, electrical properties that change over time in response to deformation of the flexible material, and the flexible material.
- the "flexible material” is a concept including a material that is at least partially deformable such as bending when an external force is applied, and has a soft elastic body such as a rubber material and a fibrous skeleton. It includes a structure and a structure in which a plurality of minute air bubbles are scattered inside. Pressure is an example of an external force. Examples of a structure having a fibrous skeleton and a structure in which a plurality of minute air bubbles are scattered inside include a polymer material such as a urethane material.
- Flexible material to which conductivity is imparted is a concept including a material having conductivity, a material to which a conductive material is imparted to a flexible material in order to impart conductivity, and a material in which a flexible material has conductivity. including.
- the flexible material to which conductivity is imparted has a function of changing the electrical characteristics according to the deformation.
- An example of a physical quantity that causes a function of changing electrical characteristics according to deformation is a pressure value due to pressure stimulation. When deforming the flexible material, it is possible to deform the flexible material with the shape of the pressure stimulus.
- an example of a physical quantity representing an electric characteristic that changes according to deformation is an electric resistance value. This electric resistance value can be regarded as the volume resistance value of the flexible material.
- the flexible material gives electrical conductivity, and the electrical characteristics according to the deformation appear. That is, as shown in FIG. 2, in the flexible material to which conductivity is imparted, the electric paths are complicatedly linked, and the electric paths expand and contract or expand and contract according to the deformation. In addition, it may show the behavior that the electric path is temporarily cut off and the behavior that a connection different from the previous one occurs.
- the flexible material exhibits different electrical properties depending on the applied force (eg, pressure stimulus) between positions separated by a predetermined distance (eg, detection points). Therefore, from the viewpoint of the shape change of the flexible material, it is considered that the electrical characteristics change according to the shape of the force applied to the flexible material (for example, pressure stimulation) and the force applied to the flexible material.
- the estimation device of the present disclosure uses the electrical characteristics associated with the time-series information that changes according to the deformation of the flexible material and the shape information of the pressure stimulus that gives the flexible material deformation as learning data. Is used as an input, and a learning model trained to output shape information is used. The estimation device inputs the electrical characteristics of the flexible material, which is the estimation target, to the learning model, and estimates the output as shape information of the estimation target.
- a conductive flexible member infiltrated into a urethane member
- a pressure stimulus by an imparting member having a predetermined shape is applied to a physical quantity that deforms the flexible material and an electric resistance value is applied to the physical quantity that changes according to the deformation of the flexible material.
- FIG. 1 shows an example of the configuration of the flexible material shape estimation device 1 as the estimation device of the present disclosure.
- the applied data (at least the shape value) labeled with the pressure stimulus given to the conductive flexible member 2 and the electric resistance data of the conductive flexible member 2 (that is, the electric resistance value). ) Is used as an input, and the shape data of the unknown pressure stimulus given to the conductive flexible member 2 is estimated and output using the trained learning model.
- the shape estimation device 1 of the flexible material estimates the shape of the pressure stimulus given to the flexible material from the electrical characteristics of the flexible material whose shape has changed due to the pressure stimulus applied to the flexible material. This makes it possible to identify the shape of the pressure stimulus given to the flexible material without using a special device or a large device or directly measuring the deformation of the flexible member.
- the conductive flexible member 2 is applied as a detection unit. That is, as shown in FIG. 1, the shape estimation device 1 for flexible materials includes an estimation unit 5. Input data 4 representing the magnitude (electrical resistance value) of the electric resistance corresponding to the pressure stimulus 3 given to the conductive flexible member 2 is input to the estimation unit 5. Further, the estimation unit 5 outputs output data 6 representing the physical quantity (shape value) of the pressure stimulus 3 given to the conductive flexible member 2 as the estimation result.
- the estimation unit 5 includes a trained learning model 51.
- the learning model 51 learns to derive the shape (output data 6) of the pressure stimulus given to the conductive flexible member 2 from the electric resistance (input data 4) of the conductive flexible member 2 to which the pressure stimulus 3 is given. It is a finished model.
- the learning model 51 is, for example, a model that defines a trained neural network, and is expressed as a set of information on the weight (strength) of the connection between the nodes (neurons) constituting the neural network.
- the learning model 51 is generated by the learning process of the learning process unit 52 (FIG. 3).
- the learning processing unit 52 performs learning processing using the time-series electrical resistance (input data 4) in the conductive flexible member 2 to which the pressure stimulus 3 is given. That is, a large amount of data obtained by measuring the electrical resistance between the detection points of the conductive flexible member 2 separated by a predetermined distance in time series using the shape of the pressure stimulus 3 as a label is used as training data.
- the training data is a large set of input data including the electric resistance value (input data 4) and information indicating the shape of the pressure stimulus 3 corresponding to the input data (output data 6). include.
- time-series information is associated with each of the electric resistance values (input data 4) of the conductive flexible member 2 by adding information indicating the measurement time.
- time-series information may be associated with the sequential set of electrical resistance values when pressure stimulation is applied to the conductive flexible member 2 by adding information indicating the measurement time.
- FIG. 4 shows an example of a measuring device 7 for measuring a physical quantity in the conductive flexible member 2.
- a pressure applying portion 73 for applying a pressure stimulus (a physical quantity that deforms the conductive flexible member 2) to the conductive flexible member 2 is attached to the fixed portion 72 fixed to the base 71.
- the pressure applying portion 73 includes a pressure applying main body 73A, an arm 73B that can be expanded and contracted from the pressure applying main body 73A, and a tip portion 73C attached to the tip of the arm 73B.
- the pressure applying main body 73A is fixed to the fixing portion 72
- the arm 73B is expanded and contracted in response to the input signal
- the tip portion 73C is moved in a predetermined direction (arrow F direction).
- the conductive flexible member 2 is installed on the base 71, and a pressing member 74 having a predetermined shape is arranged between the tip portion 73C of the pressure applying portion 73 and the conductive flexible member 2.
- the pressing members 74A to 74E shown in FIG. 5 are used as an example of the pressing member 74 having a predetermined shape.
- the pressing member 74A is a pressing member that gives a pressure stimulus in a circular shape to the conductive flexible member 2
- the pressing member 74B is a pressing member that gives a pressure stimulus in a quadrangular shape
- the pressing member 74C gives a pressure stimulus in a triangular shape. It is a pressing member to give.
- the pressing member that gives a pressure stimulus in a direction different from that of the pressing member 74B is referred to as a pressing member 74D
- the pressing member that gives a pressure stimulus in a direction different from that of the pressing member 74C is used.
- the member was a pressing member 74E.
- the pressure applying portion 73 operates so that the tip portion 73C pushes the pressing member 74 into the conductive flexible member 2 by extending the arm 73B.
- the surface of the conductive flexible member 2 on the base 71 side is provided with a detection point 75 for detecting electrical characteristics (that is, a physical quantity indicating the electrical characteristics of the conductive flexible member 2, and here, an electric resistance value).
- the detection points 75 are arranged at a plurality of different positions separated by a predetermined distance in order to detect the electric resistance value of the conductive flexible member 2.
- a plurality of detection points 75 (8 in FIG. 6) arranged in a grid pattern shown in FIG. 6 are applied as an example of the detection points 75 in the conductive flexible member 2.
- the electric resistance value for example, the volume resistance value
- FIG. 6 a reference numeral indicating each of the first to eighth points is shown in a figure (circular figure) indicating the detection point 75 as each of the eight detection points 75.
- a first detection set # 1 for detecting an electric resistance value by a first detection point 75 and a second detection point 75 is shown.
- the second detection point 75 and the third detection point 75 indicate the second detection set # 2
- the third detection point 75 and the fourth detection point 75 indicate the third detection set # 3.
- the fourth detection point 75 and the first detection point 75 indicate the fourth detection set # 4.
- the fifth detection point 75 and the seventh detection point 75 indicate the fifth detection set # 5
- the sixth detection point 75 and the eighth detection point 75 indicate the sixth detection set # 6.
- the second detection point 75 and the fourth detection point 75 indicate the seventh detection set # 7, and the first detection point 75 and the third detection point 75 indicate the eighth detection set # 8.
- the measuring device 7 includes an electrical characteristic detection unit 76 that is connected to the detection point 75 to detect electrical characteristics (that is, electric resistance value).
- the measuring device 7 includes a pressure applying unit 73 and a controller 70 connected to the electrical characteristic detecting unit 76.
- the controller 70 controls the pressure applying unit 73, gives a pressure stimulus to the conductive flexible member 2, acquires an electric resistance value due to the pressure stimulus to the conductive flexible member 2, and stores it.
- the stored electric resistance value is associated with information indicating the shape of the pressure stimulus to the conductive flexible member 2, that is, the shape of the pressing member 74.
- the measuring device 7 can acquire a plurality of data sets of electric resistance values of the conductive flexible member 2 in chronological order with respect to the shape of the pressing member 74.
- the controller 70 can be configured to include a computer including a CPU (not shown), and is adapted to execute learning data collection processing.
- FIG. 7 shows an example of the learning data collection process.
- the controller gives an instruction of pressure stimulation by the pressing member 74 to the conductive flexible member 2, and in step S102, the electric resistance value in the conductive flexible member 2 is acquired in time series.
- the shape of the pressing member 74 is assigned as a label to the acquired time-series electric resistance value and stored.
- the controller 70 is used until the shape of the pressing member 74 and the set of the electric resistance values of the conductive flexible member 2 reach a predetermined predetermined number or a predetermined predetermined time (until a positive judgment is made in step S106). Negative judgment), and repeat the above process.
- the controller 70 acquires the electric resistance value of the conductive flexible member 2 in time series for each shape of the pressing member 74 by controlling the pressing of the pressure stimulus on the conductive flexible member 2. , It becomes possible to memorize.
- the learning data is a set of electrical resistance values of the conductive flexible member 2 in time series for each shape of the pressing member 74 stored in the controller 70.
- the learning processing unit 52 includes a generator 54 and an arithmetic unit 56.
- the generator 54 has a function of generating an output in consideration of the context of the electric resistance value acquired in time series as an input.
- the learning processing unit 52 receives input data 4 (electrical resistance value) measured by the measuring device 7 as learning data, and output data 6 which is data related to the pressing member 74 given to the conductive flexible member 2 as a pressure stimulus. Holds a large number of sets with (shape).
- the generator 54 includes an input layer 540, an intermediate layer 542, and an output layer 544 to form a known neural network (NN: Neural Network). Since the neural network itself is a known technique, detailed description thereof will be omitted, but the intermediate layer 542 includes a large number of node groups (neuron groups) having inter-node connections and feedback connections.
- the data from the input layer 540 is input to the intermediate layer 542, and the data of the calculation result of the intermediate layer 542 is output to the output layer 544.
- the generator 54 is a neural network that generates generated output data 6A representing the shape of the pressing member 74 from the input input data 4 (electrical resistance).
- the generated output data 6A is data in which the shape of the pressing member 74 in which the conductive flexible member 2 is pressure-stimulated is estimated from the input data 4 (electrical resistance).
- the generator 54 generates generated output data showing a shape close to the shape of the pressing member 74 to which the conductive flexible member 2 is pressure-stimulated from the input data 4 (electrical resistance) input in time series. By learning using a large number of input data 4 (electrical resistance), the generator 54 can generate generated output data 6A that is closer to the shape of the pressing member 74 in which the conductive flexible member is pressure-stimulated. Become.
- the arithmetic unit 56 is an arithmetic unit that compares the generated output data 6A with the output data 6 of the learning data and calculates the error of the comparison result.
- the learning processing unit 52 inputs the generated output data 6A and the output data 6 of the learning data to the arithmetic unit 56. In response to this, the arithmetic unit 56 calculates an error between the generated output data 6A and the output data 6 of the learning data, and outputs a signal indicating the calculation result.
- the learning processing unit 52 learns the generator 54 that tunes the weight parameter of the connection between the nodes based on the error calculated by the arithmetic unit 56. Specifically, the weight parameter of the connection between the nodes of the input layer 540 and the intermediate layer 542 in the generator 54, the weight parameter of the connection between the nodes in the intermediate layer 542, and the node of the intermediate layer 542 and the output layer 544. Each of the weighting parameters of the coupling between them is fed back to the generator 54 by using a method such as a gradient descent method or an error back propagation method. That is, with the output data 6 of the training data as the target, the coupling between all the nodes is optimized so as to minimize the error between the output data 6A of the generated output data 6A and the output data 6 of the training data.
- the learning model 51 is generated by the learning process of the learning process unit 52.
- the learning model 51 is expressed as a set of information on weight parameters (weights or intensities) of connections between nodes of the learning result by the learning processing unit 52.
- the learning processing unit 52 can be configured to include a computer including a CPU (not shown) and execute learning processing. For example, as shown in FIG. 9 as an example of learning processing, the learning processing unit 52 uses input data labeled with information indicating the shape of the pressing member 74, which is learning data as a result of time-series measurement in step S110. Obtain 4 (electrical resistance). In step S112, the learning processing unit 52 generates a learning model 51 using the learning data of the results measured in time series. That is, a set of information on the weight parameter (weight or intensity) of the connection between the nodes of the learning result learned using a large number of training data as described above is obtained. Then, in step S114, the data represented as a set of information on the weight parameter (weight or intensity) of the connection between the nodes of the learning result is stored as the learning model 51.
- the generator 54 may use a recurrent neural network having a function of generating an output in consideration of the context of the time-series input, or may use another method.
- the trained generator 54 (that is, the data represented as a set of information of the weight parameters of the connection between the nodes of the learning result) generated by the method exemplified above is used. It is used as a learning model 51.
- the shape of the pressing member 74 pressed against the conductive flexible member 2 is identified from the time-series electrical resistance values in the pressure stimulus by the pressing member 74 against the conductive flexible member 2. It's not impossible either.
- the process by the learning process unit 52 is an example of the process of the learning model generator of the present disclosure. Further, the shape estimation device 1 of the flexible material is an example of the estimation unit and the estimation device of the present disclosure.
- the electric paths are intricately linked (see, for example, FIG. 2), and the electric paths expand / contract, expand / contract, temporarily cut, and newly connect according to the deformation. As a result, it shows behaviors having different electrical characteristics depending on a given force (for example, pressure stimulus).
- This makes it possible to treat the conductive flexible member 2 as a reservoir for storing data regarding deformation of the conductive flexible member 2. That is, the flexible material shape estimation device 1 can apply the conductive flexible member 2 to a network model (hereinafter referred to as PRCN) called physical reservoir computing (PRC). .
- PRCN network model
- PRC physical reservoir computing
- FIG. 10 shows an example of a learning processing unit 52 that learns by treating the conductive flexible member 2 as a reservoir for storing data related to deformation of the conductive flexible member 2.
- the conductive flexible member 2 has electrical characteristics (electrical resistance value) corresponding to each of various pressure stimuli, functions as an input layer for inputting an electric resistance value, and stores data on deformation of the conductive flexible member 2. Functions as a reservoir layer. Since the conductive flexible member 2 outputs different electrical characteristics (input data 4) according to the given pressure stimulus 3 (shape of the pressing member), it is given from the electric resistance value of the conductive flexible member 2 in the estimation layer. It is possible to estimate the applied pressure stimulus 3 (shape of the pressing member). Therefore, in the learning process, the estimation layer may be learned.
- the above-mentioned flexible material shape estimation device 1 can be realized by, for example, causing a computer to execute a program representing each of the above-mentioned functions.
- FIG. 11 shows an example of a case where a computer is included as an execution device that executes a process for realizing various functions of the shape estimation device 1 of a flexible material.
- the computer functioning as the shape estimation device 1 for the flexible material shown in FIG. 8 includes the computer main body 100 shown in FIG.
- the computer body 100 includes a CPU 102, a RAM 104 such as a volatile memory, a ROM 106, an auxiliary storage device 108 such as a hard disk device (HDD), and an input / output interface (I / O) 110.
- These CPU 102, RAM 104, ROM 106, auxiliary storage device 108, and input / output I / O 110 are configured to be connected via a bus 112 so that data and commands can be exchanged with each other.
- the input / output I / O 110 is connected to a communication unit 114 for communicating with an external device and an operation display unit 116 such as a display or a keyboard.
- the communication unit 114 functions to acquire input data 4 (electrical resistance) with the conductive flexible member 2. That is, the communication unit 114 includes the conductive flexible member 2 which is a detection unit, and acquires the input data 4 (electrical resistance) from the electrical characteristic detection unit 76 connected to the detection point 75 in the conductive flexible member 2. Is possible.
- the auxiliary storage device 108 stores a control program 108P for making the computer main body 100 function as the shape estimation device 1 of the flexible material as an example of the estimation device of the present disclosure.
- the CPU 102 reads the control program 108P from the auxiliary storage device 108, expands it into the RAM 104, and executes the process.
- the computer main body 100 that executes the control program 108P operates as the shape estimation device 1 of the flexible material as an example of the estimation device of the present disclosure.
- the auxiliary storage device 108 stores the learning model 108M including the learning model 51 and the data 108D including various data.
- the control program 108P may be provided by a recording medium such as a CD-ROM.
- FIG. 12 shows an example of the flow of estimation processing by the control program 108P executed in the computer main body 100.
- the estimation process shown in FIG. 12 is executed by the CPU 102 when the computer body 100 is powered on. That is, the CPU 102 reads the control program 108P from the auxiliary storage device 108, expands it into the RAM 104, and executes the process.
- step S200 the CPU 102 reads the learning model 51 from the learning model 108M of the auxiliary storage device 108 and expands it into the RAM 104 to acquire the learning model 51.
- a network model that is a connection between nodes by a weight parameter expressed as a learning model 51 is developed in the RAM 104. Therefore, the learning model 51 in which the connection between the nodes by the weight parameter is realized is constructed.
- step S202 the CPU 102 passes the unknown input data 4 (electrical resistance) to be estimated for estimating the shape of the pressing member by the pressure stimulus given to the conductive flexible member 2 via the communication unit 114. Get to the series.
- unknown input data 4 electrical resistance
- step S204 the CPU 102 uses the learning model 51 acquired in step S200 to obtain output data 6 (shape of an unknown estimated object) corresponding to the input data 4 (electrical resistance) acquired in step S202. presume.
- the output data 6 (shape of the estimation target object) of the estimation result is output via the communication unit 114, and this processing routine is terminated.
- the estimation process shown in FIG. 12 is an example of the process executed by the estimation method of the present disclosure.
- the conductive flexible member 2 is subjected to a pressure stimulus from the input data 4 (electrical resistance) that changes in response to the given pressure stimulus 3. It becomes possible to estimate the shape of a given object. That is, it is possible to estimate the pressure stimulus applied to the flexible member, for example, the shape of the estimation target, without using a special device or a large device or directly measuring the deformation of the flexible member.
- FIG. 13A to 13D show the characteristics of the pressure stimulus and the electrical characteristics (electrical resistance value) of the conductive flexible member 2 when the pressure stimulus is applied to the conductive flexible member 2 by the pressing member 74. ..
- FIG. 13A shows the time characteristic of the pressure value which is the pressure stimulus applied to the conductive flexible member 2.
- FIG. 13B shows the time characteristic of the depth of the deformation amount of the conductive flexible member 2 when the pressure value shown in FIG. 13A is applied to the conductive flexible member 2.
- FIG. 13C shows the time characteristics of the electrical resistance value of the conductive flexible member 2 in the detection set # 1 when the pressure value shown in FIG. 13A is given to the conductive flexible member 2.
- FIG. 13D shows the time characteristic regarding the start of pressure stimulation on the conductive flexible member 2, that is, the amount of change in the electric resistance value from the start of pressing.
- An example of the time characteristic regarding the amount of change in the electric resistance value is the time characteristic of the difference.
- the electric resistance value (volume resistance value) of the conductive flexible member 2 changes in response to the pressure stimulus. Can be confirmed to indicate. Further, it can be confirmed from the time characteristic (for example, the difference time characteristic indicating the difference) regarding the amount of change in the electric resistance value from the start of pressing that the electric resistance value changes in time series according to the deformation of the flexible material.
- FIG. 14 shows the verification result of verifying the estimation result due to the difference in the shape of the pressing member 74 that gives a pressure stimulus to the conductive flexible member 2.
- the pressing members 74 were subjected to 40 pressing tests for each of the pressing members 74A to 74E having different shapes, and the electric resistance values changed in each time series were collected as learning data. Then, a learning model learned from the collected learning data was constructed, and the shape estimation test was carried out 30 times using the learning model.
- the pressing member 74A is referred to as C 0
- the pressing member 74B is referred to as S 0
- the pressing member 74C is referred to as T 0
- the pressing member 74D is referred to as T 90
- the pressing member 74E is referred to as T 90. It is written as S 45 .
- the estimation result shows the estimated number of times for the shape of the estimated member when the pressing member 74 is pressed.
- the shape can be estimated even when a pressure stimulus is applied to the conductive flexible member 2 by a pressing member 74 having a different shape. Can be confirmed as.
- FIG. 15 shows the verification result of verifying the estimation result of the shape of the pressing member 74 with different detection sets.
- the same pressing member 74 was subjected to 40 pressing tests for each of the different detection sets # 1 to # 8, and the electric resistance values changed in each time series were collected as learning data. Then, a learning model learned from the collected learning data was constructed, and the shape estimation test was carried out 30 times using the learning model.
- a detection set that is, a detection position (for example, detection set # 1)
- a detection position for example, detection set # 1
- FIG. 16 shows the verification result of verifying the estimation result of the shape of the pressing member 74 by the combination of a plurality of different detection sets.
- the above-mentioned 40 times of pressing test was performed, and 30 times of shape estimation test was carried out using the learning model learned from the collected learning data.
- FIG. 17A and 17B show time characteristics regarding the electrical characteristics (electrical resistance value) of the conductive flexible member 2 when a pressure stimulus is applied to the conductive flexible member 2 by each of the plurality of pressing members 74.
- FIG. 17A shows the time characteristics of the electric resistance value of the conductive flexible member 2 when a pressure stimulus is applied to the conductive flexible member 2 in each of the pressing members 74A to 74E.
- FIG. 17B shows the amount of change in the electric resistance value from the start of pressure stimulation, that is, the start of pressing, for each of the pressing members 74A to 74E.
- FIGS. 17A and 17B show the results of three-dimensional display of the results of principal component analysis with respect to the electrical characteristics shown in FIGS. 17A and 17B.
- FIG. 18A shows the principal component analysis results for each of the pressing members 74A to 74E
- FIG. 18B shows the principal component analysis results for the pressing members 74D and 74E.
- each of the pressing members 74A to 74E having different shapes has a separated structure, and it can be confirmed that each of the pressing members 74A to 74E having different shapes is different.
- FIG. 19 shows the verification result of verifying the estimation result due to the difference in the size and the pressing position of the pressing member 74.
- the pressing member 74A according to the condition of being separated from the learned position of the circular pressing member 74A by a predetermined distance (for example, 5 mm) is referred to as C1, and the condition of the position further separated (for example, 10 mm).
- the pressing member 74A by the above is referred to as C 2 .
- the pressing member 74A under the condition (for example, diameter 40 mm) reduced from the size (for example, diameter 50 mm) of the pressing member 74A at the time of learning is referred to as C3 , and the pressing member 74A under the condition further reduced (for example, diameter 30 mm).
- FIG. 20 shows the verification result of verifying the estimation result using the learning model learned under various conditions due to the difference in the size and the pressing position of the pressing member 74 described above. As shown in FIG. 20, it can be confirmed that the shape can be estimated satisfactorily under almost all conditions.
- FIG. 21A and 21B show the results of three-dimensional display of the results of principal component analysis of the verification results shown in FIG. 20.
- FIG. 21A shows the result of detecting the electric resistance value by the detection set # 1
- FIG. 21B shows the result of referring to FIG. 21A from different directions.
- the first aspect is A detector that detects electrical properties between multiple detection points in a flexible material with conductivity, The electrical characteristics are input by using the electrical characteristics that change in time series according to the deformation of the flexible material and the shape information indicating the shape of the flexible material in the pressure stimulus that deforms the flexible material as learning data. Then, to the learning model trained to output the shape information, an estimation unit that inputs the electrical characteristics detected by the detection unit of the estimation target and estimates the shape information of the estimation target, and an estimation unit. It is an estimation device including.
- the second aspect is in the estimation device of the first aspect.
- the flexible material is a material whose electrical characteristics change in response to the deformation.
- the learning model is trained to output shape information corresponding to the detected electrical characteristics.
- the third aspect is in the estimation device of the first aspect or the second aspect.
- the electrical property of the flexible material is volume resistance.
- the fourth aspect is the estimation device of any one of the first to third aspects.
- the flexible material is a material in which conductivity is imparted to a urethane material having a structure having a fibrous skeleton or a structure in which a plurality of minute air bubbles are scattered inside.
- the fifth aspect is in the estimation device of any one of the first to fourth aspects.
- the learning model is a model generated by learning using the flexible material as a reservoir using a network by reservoir computing using the reservoir.
- the sixth aspect is The computer acquires the electrical properties from the detector that detects the electrical properties between multiple detection points in the flexible material having conductivity. Using the electrical characteristics associated with the time-series information that changes according to the deformation of the flexible material and the shape information of the pressure stimulus that gives deformation to the flexible material as learning data, the electrical characteristics are input. The acquired electrical characteristics of the estimation target are input to the learning model trained to output the shape information, and the shape information of the estimation target is estimated. It is an estimation method.
- the seventh aspect is Obtaining the electrical properties from a detector that detects the electrical properties between multiple detection points in a flexible material that has conductivity to a computer, Using the electrical characteristics associated with the time-series information that changes according to the deformation of the flexible material and the shape information of the pressure stimulus that gives deformation to the flexible material as learning data, the electrical characteristics are input.
- the acquired electrical characteristics of the estimation target are input to the learning model trained to output the shape information, and the shape information of the estimation target is estimated. It is an estimation program for executing processing.
- the eighth aspect is An acquisition unit that acquires the electrical characteristics from the detection unit that detects the electrical characteristics between a plurality of detection points in the flexible material having conductivity, and the shape information of the pressure stimulus that gives deformation to the flexible material. Based on the acquisition result of the acquisition unit, a learning model trained to output the shape information of the object by inputting the electrical characteristics associated with the time-series information that changes according to the deformation of the flexible material.
- the learning model generator to generate and It is a learning model generator including.
- the shape information of the pressure stimulus can be estimated by utilizing the electrical characteristics at the time of deformation of the flexible material having conductivity without using a special detection device.
- the conductive flexible member is applied as an example of the flexible member
- the flexible member is not limited to the conductive flexible member.
- a part of the estimation device for example, a neural network such as a learning model may be configured as a hardware circuit.
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Force Measurement Appropriate To Specific Purposes (AREA)
- Measurement Of Length, Angles, Or The Like Using Electric Or Magnetic Means (AREA)
Abstract
Description
なお、作用、機能が同じ働きを担う構成要素及び処理には、全図面を通して同じ符合を付与し、重複する説明を適宜省略する場合がある。また、本開示は、以下の実施形態に何ら限定されるものではなく、本開示の目的の範囲内において、適宜変更を加えて実施することができる。また、本開示では、主として非線形に変形する部材に対する物理量の推定を説明するが、線形に変形する部材に対する物理量の推定に適用可能であることは言うまでもない。
図4に、導電性柔軟部材2における物理量を測定する測定装置7の一例を示す。
学習処理部52は、生成器54と演算器56とを含む。生成器54は、入力である時系列に取得された電気抵抗値の前後関係を考慮して出力を生成する機能を有する。
図12に示す推定処理は、コンピュータ本体100に電源投入されると、CPU102により実行される。すなわち、CPU102は、制御プログラム108Pを補助記憶装置108から読み出し、RAM104に展開して処理を実行する。
導電性を有する柔軟材料における複数の検出点の間の電気特性を検出する検出部と、
前記柔軟材料の変形に応じて時系列に変化する電気特性と、前記柔軟材料に変形を与える圧力刺激の前記柔軟材料における形状を示す形状情報とを学習用データとして用いて、前記電気特性を入力とし、前記形状情報を出力するように学習された学習モデルに対して、推定対象物の前記検出部で検出された電気特性を入力し、前記推定対象物の形状情報を推定する推定部と、
を含む推定装置である。
前記柔軟材料は、前記変形に応じて電気特性が変化する材料であり、
前記学習モデルは、検出された電気特性に対応する形状情報を出力するように学習される。
前記柔軟材料の電気特性は、体積抵抗である。
前記柔軟材料は、繊維状の骨格を有する構造、又は内部に微小な空気泡が複数散在する構造のウレタン材に導電性が付与された材料である。
前記学習モデルは、前記柔軟材料をリザーバとして当該リザーバを用いたリザーバコンピューティングによるネットワークを用いて学習させることで生成されたモデルである。
コンピュータが
導電性を有する柔軟材料における複数の検出点の間の電気特性を検出する検出部からの前記電気特性を取得し、
前記柔軟材料の変形に応じて変化する時系列情報が対応付けられた電気特性と、前記柔軟材料に変形を与える圧力刺激の形状情報とを学習用データとして用いて、前記電気特性を入力とし、前記形状情報を出力するように学習された学習モデルに対して、推定対象物の前記取得された電気特性を入力し、前記推定対象物の形状情報を推定する、
推定方法である。
コンピュータに
導電性を有する柔軟材料における複数の検出点の間の電気特性を検出する検出部からの前記電気特性を取得し、
前記柔軟材料の変形に応じて変化する時系列情報が対応付けられた電気特性と、前記柔軟材料に変形を与える圧力刺激の形状情報とを学習用データとして用いて、前記電気特性を入力とし、前記形状情報を出力するように学習された学習モデルに対して、推定対象物の前記取得された電気特性を入力し、前記推定対象物の形状情報を推定する、
処理を実行させるための推定プログラムである。
導電性を有する柔軟材料における複数の検出点の間の電気特性を検出する検出部からの前記電気特性と、前記柔軟材料に変形を与える圧力刺激の形状情報と、を取得する取得部と、
前記取得部の取得結果に基づいて、前記柔軟材料の変形に応じて変化する時系列情報が対応付けられた電気特性を入力とし、対象物の形状情報を出力するように学習された学習モデルを生成する学習モデル生成部と、
を含む学習モデル生成装置である。
Claims (8)
- 導電性を有する柔軟材料における複数の検出点の間の電気特性を検出する検出部と、
前記柔軟材料の変形に応じて時系列に変化する電気特性と、前記柔軟材料に変形を与える圧力刺激の前記柔軟材料における形状を示す形状情報とを学習用データとして用いて、前記電気特性を入力とし、前記形状情報を出力するように学習された学習モデルに対して、推定対象物の前記検出部で検出された電気特性を入力し、前記推定対象物の形状情報を推定する推定部と、
を含む推定装置。 - 前記柔軟材料は、前記変形に応じて電気特性が変化する材料であり、
前記学習モデルは、検出された電気特性に対応する形状情報を出力するように学習される
請求項1に記載の推定装置。 - 前記柔軟材料の電気特性は、体積抵抗である、
請求項1又は請求項2に記載の推定装置。 - 前記柔軟材料は、繊維状の骨格を有する構造、又は内部に微小な空気泡が複数散在する構造のウレタン材に導電性が付与された材料である、
請求項1から請求項3の何れか1項に記載の推定装置。 - 前記学習モデルは、前記柔軟材料をリザーバとして当該リザーバを用いたリザーバコンピューティングによるネットワークを用いて学習させることで生成されたモデルである
請求項1から請求項4の何れか1項に記載の推定装置。 - コンピュータが
導電性を有する柔軟材料における複数の検出点の間の電気特性を検出する検出部からの前記電気特性を取得し、
前記柔軟材料の変形に応じて変化する時系列情報が対応付けられた電気特性と、前記柔軟材料に変形を与える圧力刺激の形状情報とを学習用データとして用いて、前記電気特性を入力とし、前記形状情報を出力するように学習された学習モデルに対して、推定対象物の前記取得された電気特性を入力し、前記推定対象物の形状情報を推定する、
推定方法。 - コンピュータに
導電性を有する柔軟材料における複数の検出点の間の電気特性を検出する検出部からの前記電気特性を取得し、
前記柔軟材料の変形に応じて変化する時系列情報が対応付けられた電気特性と、前記柔軟材料に変形を与える圧力刺激の形状情報とを学習用データとして用いて、前記電気特性を入力とし、前記形状情報を出力するように学習された学習モデルに対して、推定対象物の前記取得された電気特性を入力し、前記推定対象物の形状情報を推定する、
処理を実行させるための推定プログラム。 - 導電性を有する柔軟材料における複数の検出点の間の電気特性を検出する検出部からの前記電気特性と、前記柔軟材料に変形を与える圧力刺激の形状情報と、を取得する取得部と、
前記取得部の取得結果に基づいて、前記柔軟材料の変形に応じて変化する時系列情報が対応付けられた電気特性を入力とし、対象物の形状情報を出力するように学習された学習モデルを生成する学習モデル生成部と、
を含む学習モデル生成装置。
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US18/257,101 US20240104374A1 (en) | 2020-12-18 | 2021-12-09 | Estimation device, estimation method, estimation program, and learning model generation device |
CN202180084131.0A CN116583710A (zh) | 2020-12-18 | 2021-12-09 | 估计装置、估计方法、估计程序以及学习模型生成装置 |
EP21906501.8A EP4265999A4 (en) | 2020-12-18 | 2021-12-09 | ESTIMATION DEVICE, ESTIMATION METHOD, ESTIMATION PROGRAM AND LEARNING MODEL CREATION DEVICE |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2020-210737 | 2020-12-18 | ||
JP2020210737A JP7487887B2 (ja) | 2020-12-18 | 2020-12-18 | 推定装置、推定方法、推定プログラム、及び学習モデル生成装置 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022131138A1 true WO2022131138A1 (ja) | 2022-06-23 |
Family
ID=82059150
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2021/045417 WO2022131138A1 (ja) | 2020-12-18 | 2021-12-09 | 推定装置、推定方法、推定プログラム、及び学習モデル生成装置 |
Country Status (5)
Country | Link |
---|---|
US (1) | US20240104374A1 (ja) |
EP (1) | EP4265999A4 (ja) |
JP (1) | JP7487887B2 (ja) |
CN (1) | CN116583710A (ja) |
WO (1) | WO2022131138A1 (ja) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2013101096A (ja) | 2011-10-11 | 2013-05-23 | Touchence Inc | 柔軟触覚センサ |
JP2014228507A (ja) * | 2013-05-27 | 2014-12-08 | 日本電信電話株式会社 | 伸長センサおよび測定装置 |
WO2017029905A1 (ja) | 2015-08-19 | 2017-02-23 | 国立研究開発法人産業技術総合研究所 | 単一カメラによる物体の変位と振動の測定方法、装置およびそのプログラム |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
FR3088006B1 (fr) * | 2018-11-06 | 2022-12-16 | Inria Inst Nat Rech Informatique & Automatique | Textile intelligent adapte pour la detection de mouvement et/ou de deformation |
-
2020
- 2020-12-18 JP JP2020210737A patent/JP7487887B2/ja active Active
-
2021
- 2021-12-09 EP EP21906501.8A patent/EP4265999A4/en active Pending
- 2021-12-09 WO PCT/JP2021/045417 patent/WO2022131138A1/ja active Application Filing
- 2021-12-09 CN CN202180084131.0A patent/CN116583710A/zh active Pending
- 2021-12-09 US US18/257,101 patent/US20240104374A1/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2013101096A (ja) | 2011-10-11 | 2013-05-23 | Touchence Inc | 柔軟触覚センサ |
JP2014228507A (ja) * | 2013-05-27 | 2014-12-08 | 日本電信電話株式会社 | 伸長センサおよび測定装置 |
WO2017029905A1 (ja) | 2015-08-19 | 2017-02-23 | 国立研究開発法人産業技術総合研究所 | 単一カメラによる物体の変位と振動の測定方法、装置およびそのプログラム |
Non-Patent Citations (2)
Title |
---|
NAKAJIMA, KOHEI ET AL.: "Exploiting the Deformation Dynamics of Soft Materials as an Information Processing Device", THE TRANSACTIONS OF THE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS, vol. 102, no. 2, 1 February 2019 (2019-02-01), pages 121 - 126, XP009537488 * |
See also references of EP4265999A4 |
Also Published As
Publication number | Publication date |
---|---|
JP7487887B2 (ja) | 2024-05-21 |
CN116583710A (zh) | 2023-08-11 |
EP4265999A1 (en) | 2023-10-25 |
US20240104374A1 (en) | 2024-03-28 |
JP2022097262A (ja) | 2022-06-30 |
EP4265999A4 (en) | 2024-06-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Park et al. | Monitoring impact events using a system-identification method | |
JP7432201B2 (ja) | 推定装置、推定方法、プログラム、及び学習モデル生成装置 | |
CN106204620A (zh) | 一种基于微视觉的触觉三维力检测方法 | |
Borate et al. | Data-driven structural health monitoring approach using guided Lamb wave responses | |
KR20210035630A (ko) | 유연촉각센서, 유연촉각센서 시스템 및 그 제조방법 | |
JP6232273B2 (ja) | 指先接触状態測定装置 | |
Park et al. | Softness-adaptive pinch-grasp strategy using fingertip tactile information of robot hand | |
WO2022131138A1 (ja) | 推定装置、推定方法、推定プログラム、及び学習モデル生成装置 | |
Gurung et al. | Self-sensing SMA actuator using extended Kalman filter and artificial neural network | |
JP7525441B2 (ja) | 推定装置、推定方法、及びプログラム | |
Khalil et al. | Sensor measurements only with transmissibility operators for fault detection in flexible beams | |
JPWO2020170770A1 (ja) | 検出装置及びセンサのキャリブレーション方法 | |
JP2023017635A (ja) | 推定装置、推定方法、推定プログラム、ロボットシステム及び学習モデル生成装置 | |
Sladek et al. | The dynamic response of a tactile sensor | |
Yan et al. | Computational framework for dense sensor network evaluation based on model-assisted probability of detection | |
WO2023112368A1 (ja) | 推定装置、推定方法、推定プログラム、及び学習モデル生成装置 | |
JP2023088197A (ja) | 推定装置、推定方法、推定プログラム、及び学習モデル生成装置 | |
WO2023182012A1 (ja) | 推定装置、推定方法、推定プログラム、及び学習モデル生成装置 | |
Zhang | Simulation of the SynTouch BioTac Sensor | |
Yousif et al. | Parameter Identification in Elasto-Plastic Plates using Neural Networks | |
JP2023017636A (ja) | 推定装置、推定方法、推定プログラム、ロボットシステム及び学習モデル生成装置 | |
JP2023088171A (ja) | 推定装置、推定方法、推定プログラム、及び学習モデル生成装置 | |
JP2023088191A (ja) | 推定装置、推定方法、推定プログラム、及び学習モデル生成装置 | |
JP2023088188A (ja) | 推定装置、推定方法、推定プログラム、及び学習モデル生成装置 | |
JP2023017631A (ja) | 推定装置、推定方法、推定プログラム、及び学習モデル生成装置 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 21906501 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 18257101 Country of ref document: US |
|
WWE | Wipo information: entry into national phase |
Ref document number: 202180084131.0 Country of ref document: CN |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
ENP | Entry into the national phase |
Ref document number: 2021906501 Country of ref document: EP Effective date: 20230718 |